In [ ]:
cv2_imshow(images[1])
In recent times, the field of agriculture has been in urgent need of modernizing, since the amount of manual work people need to put in to check if plants are growing correctly is still highly extensive. Despite several advances in agricultural technology, people working in the agricultural industry still need to have the ability to sort and recognize different plants and weeds, which takes a lot of time and effort in the long term. The potential is ripe for this trillion-dollar industry to be greatly impacted by technological innovations that cut down on the requirement for manual labor, and this is where Artificial Intelligence can actually benefit the workers in this field, as the time and energy required to identify plant seedlings will be greatly shortened by the use of AI and Deep Learning. The ability to do so far more efficiently and even more effectively than experienced manual labor, could lead to better crop yields, the freeing up of human inolvement for higher-order agricultural decision making, and in the long term will result in more sustainable environmental practices in agriculture as well.
The aim of this project is to Build a Convolutional Neural Netowrk to classify plant seedlings into their respective categories.
The Aarhus University Signal Processing group, in collaboration with the University of Southern Denmark, has recently released a dataset containing images of unique plants belonging to 12 different species.
Due to the large volume of data, the images were converted to the images.npy file and the labels are also put into Labels.csv, so that you can work on the data/project seamlessly without having to worry about the high data volume.
The goal of the project is to create a classifier capable of determining a plant's species from an image.
List of Species
import os
import numpy as np # Importing numpy for Matrix Operations
import pandas as pd # Importing pandas to read CSV files
import matplotlib.pyplot as plt # Importting matplotlib for Plotting and visualizing images
import math # Importing math module to perform mathematical operations
import cv2 # Importing openCV for image processing
import seaborn as sns # Importing seaborn to plot graphs
# Tensorflow modules
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator # Importing the ImageDataGenerator for data augmentation
from tensorflow.keras.models import Sequential, Model # Importing the sequential module to define a sequential model
from tensorflow.keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPooling2D,BatchNormalization # Defining all the layers to build our CNN Model
from tensorflow.keras.optimizers import Adam,SGD # Importing the optimizers which can be used in our model
from sklearn import preprocessing # Importing the preprocessing module to preprocess the data
from sklearn.model_selection import train_test_split # Importing train_test_split function to split the data into train and test
from sklearn.metrics import confusion_matrix, accuracy_score, roc_auc_score, roc_curve # Importing confusion_matrix to plot the confusion matrix
from sklearn.preprocessing import LabelBinarizer
# Display images using OpenCV
from google.colab.patches import cv2_imshow # Importing cv2_imshow from google.patches to display images
from sklearn.model_selection import train_test_split
from tensorflow.keras import backend
from keras.callbacks import ReduceLROnPlateau
import random
# Ignore warnings
import warnings
warnings.filterwarnings('ignore')
# Mount Google drive to access the dataset
# Run the below code if you using google colab
from google.colab import drive
drive.mount('/content/drive')
Mounted at /content/drive
# Load the image file of dataset
images = np.load('/content/drive/MyDrive/Python | AI | ML/Colab Notebooks/Computer Vision/Project/Plant_Images.npy')
# Load the labels file of dataset
labels = pd.read_csv('/content/drive/MyDrive/Python | AI | ML/Colab Notebooks/Computer Vision/Project/Plant_Labels.csv')
print(images.shape)
print(labels.shape)
(4750, 128, 128, 3) (4750, 1)
cv2_imshow(images[1])
plt.imshow(images[1])
<matplotlib.image.AxesImage at 0x7de5124eaa40>
# Converting the images from BGR to RGB using cvtColor function of OpenCV
for i in range(len(images)):
images[i] = cv2.cvtColor(images[i], cv2.COLOR_BGR2RGB)
def plot_images(images,labels):
num_classes=10 # Number of Classes
categories=np.unique(labels)
keys=dict(labels['Label']) # Obtaing the unique classes from y_train
rows = 3 # Defining number of rows=3
cols = 4 # Defining number of columns=4
fig = plt.figure(figsize=(10, 8)) # Defining the figure size to 10x8
for i in range(cols):
for j in range(rows):
random_index = np.random.randint(0, len(labels)) # Generating random indices from the data and plotting the images
ax = fig.add_subplot(rows, cols, i * rows + j + 1) # Adding subplots with 3 rows and 4 columns
ax.imshow(images[random_index, :]) # Plotting the image
ax.set_title(keys[random_index])
plt.show()
plot_images(images,labels)
def plot_images_unique(images, labels):
num_classes = 10 # Number of Classes
categories = np.unique(labels)
keys = labels['Label'].unique() # Obtaining the unique classes from the 'Label' column
rows = 3 # Defining number of rows=3
cols = 4 # Defining number of columns=4
fig, axes = plt.subplots(rows, cols, figsize=(10, 8)) # Defining the figure size to 10x8
for i, category in enumerate(categories):
# Find the indices of images with the current category
category_indices = [index for index, label in enumerate(labels['Label']) if label == category]
# Plot an image for the current category
ax = axes[i // cols, i % cols]
ax.imshow(images[category_indices[0], :]) # Display the first image of the category
ax.set_title(str(category))
# Hide empty subplots
for i in range(len(categories), rows * cols):
fig.delaxes(axes.flatten()[i])
plt.show()
plot_images_unique(images,labels)
print(labels['Label'].dtype)
object
labels['Label'] = labels['Label'].astype(str)
plt.figure(figsize=(10, 6))
sns.countplot(x='Label', data=labels, palette='Spectral')
plt.xticks(rotation='vertical')
plt.show()
Shepherds Purse, Common Wheat, Cleavers, Maize, and Black-Grass.Split the dataset
from sklearn.model_selection import train_test_split
X_temp, X_test, y_temp, y_test = train_test_split(np.array(images),labels , test_size=0.1, random_state=42,stratify=labels)
X_train, X_val, y_train, y_val = train_test_split(X_temp,y_temp , test_size=0.1, random_state=42,stratify=y_temp)
print(X_train.shape,y_train.shape)
print(X_val.shape,y_val.shape)
print(X_test.shape,y_test.shape)
(3847, 128, 128, 3) (3847, 1) (428, 128, 128, 3) (428, 1) (475, 128, 128, 3) (475, 1)
# Convert labels from names to one hot vectors.
# We have already used encoding methods like onehotencoder and labelencoder earlier so now we will be using a new encoding method called labelBinarizer.
# Labelbinarizer works similar to onehotencoder
from sklearn.preprocessing import LabelBinarizer
enc = LabelBinarizer()
y_train_encoded = enc.fit_transform(y_train)
y_val_encoded=enc.transform(y_val)
y_test_encoded=enc.transform(y_test)
# Normalizing the image pixels
X_train_normalized = X_train.astype('float32')/255.0
X_val_normalized = X_val.astype('float32')/255.0
X_test_normalized = X_test.astype('float32')/255.0
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model1 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model1.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
# Adding max pooling to reduce the size of output of first conv layer
model1.add(MaxPooling2D((2, 2), padding = 'same'))
model1.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model1.add(MaxPooling2D((2, 2), padding = 'same'))
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model1.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model1.add(Dense(16, activation='relu'))
model1.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model1.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model1.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model1.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
max_pooling2d (MaxPooling2 (None, 64, 64, 64) 0
D)
conv2d_1 (Conv2D) (None, 64, 64, 32) 18464
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 544764 (2.08 MB)
Trainable params: 544764 (2.08 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
history1 = model1.fit(
X_train_normalized, y_train_encoded,
epochs=30,
validation_data=(X_val_normalized,y_val_encoded),
batch_size=32,
verbose=2
)
Epoch 1/30 121/121 - 16s - loss: 2.4219 - accuracy: 0.1586 - val_loss: 2.3154 - val_accuracy: 0.2266 - 16s/epoch - 128ms/step Epoch 2/30 121/121 - 2s - loss: 2.2147 - accuracy: 0.2826 - val_loss: 2.1083 - val_accuracy: 0.3271 - 2s/epoch - 18ms/step Epoch 3/30 121/121 - 2s - loss: 2.0983 - accuracy: 0.3018 - val_loss: 2.0678 - val_accuracy: 0.3481 - 2s/epoch - 17ms/step Epoch 4/30 121/121 - 2s - loss: 2.0260 - accuracy: 0.3140 - val_loss: 1.9203 - val_accuracy: 0.3668 - 2s/epoch - 17ms/step Epoch 5/30 121/121 - 2s - loss: 2.0010 - accuracy: 0.3221 - val_loss: 1.8672 - val_accuracy: 0.3762 - 2s/epoch - 17ms/step Epoch 6/30 121/121 - 2s - loss: 1.9615 - accuracy: 0.3280 - val_loss: 1.8477 - val_accuracy: 0.3762 - 2s/epoch - 17ms/step Epoch 7/30 121/121 - 2s - loss: 1.9560 - accuracy: 0.3254 - val_loss: 1.8176 - val_accuracy: 0.3925 - 2s/epoch - 18ms/step Epoch 8/30 121/121 - 2s - loss: 1.9050 - accuracy: 0.3275 - val_loss: 1.7660 - val_accuracy: 0.3832 - 2s/epoch - 19ms/step Epoch 9/30 121/121 - 2s - loss: 1.8985 - accuracy: 0.3332 - val_loss: 1.7665 - val_accuracy: 0.3972 - 2s/epoch - 19ms/step Epoch 10/30 121/121 - 2s - loss: 1.8558 - accuracy: 0.3460 - val_loss: 1.7090 - val_accuracy: 0.3972 - 2s/epoch - 17ms/step Epoch 11/30 121/121 - 2s - loss: 1.8346 - accuracy: 0.3434 - val_loss: 1.6590 - val_accuracy: 0.4182 - 2s/epoch - 17ms/step Epoch 12/30 121/121 - 3s - loss: 1.8187 - accuracy: 0.3494 - val_loss: 1.6596 - val_accuracy: 0.4065 - 3s/epoch - 23ms/step Epoch 13/30 121/121 - 2s - loss: 1.8089 - accuracy: 0.3429 - val_loss: 1.6524 - val_accuracy: 0.4042 - 2s/epoch - 20ms/step Epoch 14/30 121/121 - 2s - loss: 1.8048 - accuracy: 0.3392 - val_loss: 1.6473 - val_accuracy: 0.4136 - 2s/epoch - 18ms/step Epoch 15/30 121/121 - 2s - loss: 1.7803 - accuracy: 0.3449 - val_loss: 1.6684 - val_accuracy: 0.3925 - 2s/epoch - 18ms/step Epoch 16/30 121/121 - 2s - loss: 1.7825 - accuracy: 0.3535 - val_loss: 1.6707 - val_accuracy: 0.4065 - 2s/epoch - 19ms/step Epoch 17/30 121/121 - 2s - loss: 1.7610 - accuracy: 0.3486 - val_loss: 1.5750 - val_accuracy: 0.4112 - 2s/epoch - 17ms/step Epoch 18/30 121/121 - 2s - loss: 1.7596 - accuracy: 0.3470 - val_loss: 1.5803 - val_accuracy: 0.4276 - 2s/epoch - 17ms/step Epoch 19/30 121/121 - 2s - loss: 1.7353 - accuracy: 0.3621 - val_loss: 1.5534 - val_accuracy: 0.4299 - 2s/epoch - 17ms/step Epoch 20/30 121/121 - 2s - loss: 1.7240 - accuracy: 0.3647 - val_loss: 1.5911 - val_accuracy: 0.4136 - 2s/epoch - 17ms/step Epoch 21/30 121/121 - 2s - loss: 1.7151 - accuracy: 0.3569 - val_loss: 1.5684 - val_accuracy: 0.4276 - 2s/epoch - 18ms/step Epoch 22/30 121/121 - 2s - loss: 1.7104 - accuracy: 0.3644 - val_loss: 1.5922 - val_accuracy: 0.4159 - 2s/epoch - 18ms/step Epoch 23/30 121/121 - 2s - loss: 1.6917 - accuracy: 0.3717 - val_loss: 1.5872 - val_accuracy: 0.4463 - 2s/epoch - 19ms/step Epoch 24/30 121/121 - 2s - loss: 1.6921 - accuracy: 0.3759 - val_loss: 1.5355 - val_accuracy: 0.4416 - 2s/epoch - 18ms/step Epoch 25/30 121/121 - 2s - loss: 1.6789 - accuracy: 0.3767 - val_loss: 1.5048 - val_accuracy: 0.4673 - 2s/epoch - 17ms/step Epoch 26/30 121/121 - 2s - loss: 1.6359 - accuracy: 0.4053 - val_loss: 1.5006 - val_accuracy: 0.4977 - 2s/epoch - 17ms/step Epoch 27/30 121/121 - 2s - loss: 1.6080 - accuracy: 0.4081 - val_loss: 1.5291 - val_accuracy: 0.4836 - 2s/epoch - 17ms/step Epoch 28/30 121/121 - 2s - loss: 1.5977 - accuracy: 0.4214 - val_loss: 1.4772 - val_accuracy: 0.5164 - 2s/epoch - 17ms/step Epoch 29/30 121/121 - 2s - loss: 1.5504 - accuracy: 0.4318 - val_loss: 1.4837 - val_accuracy: 0.5070 - 2s/epoch - 19ms/step Epoch 30/30 121/121 - 2s - loss: 1.5718 - accuracy: 0.4276 - val_loss: 1.5596 - val_accuracy: 0.4883 - 2s/epoch - 19ms/step
plt.plot(history1.history['accuracy'])
plt.plot(history1.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model1.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 1.6333 - accuracy: 0.5032 - 284ms/epoch - 19ms/step
# Here we would get the output as probablities for each category
y_pred=model1.predict(X_test_normalized)
15/15 [==============================] - 0s 8ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model2 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model2.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model2.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model2.add(MaxPooling2D((2, 2), padding = 'same'))
model2.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model2.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model2.add(MaxPooling2D((2, 2), padding = 'same'))
model2.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model2.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model2.add(Dense(16, activation='relu'))
model2.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model2.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model2.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model2.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
conv2d_1 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d (MaxPooling2 (None, 64, 64, 32) 0
D)
conv2d_2 (Conv2D) (None, 64, 64, 32) 9248
conv2d_3 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization (Batch (None, 32, 32, 32) 128
Normalization)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 563388 (2.15 MB)
Trainable params: 563324 (2.15 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
history2 = model2.fit(
X_train_normalized, y_train_encoded,
epochs=30,
validation_data=(X_val_normalized,y_val_encoded),
batch_size=32,
verbose=2
)
Epoch 1/30 121/121 - 8s - loss: 2.2222 - accuracy: 0.2394 - val_loss: 2.4101 - val_accuracy: 0.1963 - 8s/epoch - 64ms/step Epoch 2/30 121/121 - 4s - loss: 1.7732 - accuracy: 0.3660 - val_loss: 2.2091 - val_accuracy: 0.3364 - 4s/epoch - 33ms/step Epoch 3/30 121/121 - 4s - loss: 1.4928 - accuracy: 0.4630 - val_loss: 2.2885 - val_accuracy: 0.2103 - 4s/epoch - 32ms/step Epoch 4/30 121/121 - 4s - loss: 1.3418 - accuracy: 0.5152 - val_loss: 1.8699 - val_accuracy: 0.3224 - 4s/epoch - 32ms/step Epoch 5/30 121/121 - 4s - loss: 1.2341 - accuracy: 0.5545 - val_loss: 1.1352 - val_accuracy: 0.6519 - 4s/epoch - 33ms/step Epoch 6/30 121/121 - 4s - loss: 1.1078 - accuracy: 0.6075 - val_loss: 1.7773 - val_accuracy: 0.4696 - 4s/epoch - 33ms/step Epoch 7/30 121/121 - 4s - loss: 1.0226 - accuracy: 0.6197 - val_loss: 1.9402 - val_accuracy: 0.4650 - 4s/epoch - 33ms/step Epoch 8/30 121/121 - 4s - loss: 0.9205 - accuracy: 0.6681 - val_loss: 1.1610 - val_accuracy: 0.6425 - 4s/epoch - 32ms/step Epoch 9/30 121/121 - 4s - loss: 0.8198 - accuracy: 0.7039 - val_loss: 1.8444 - val_accuracy: 0.5070 - 4s/epoch - 32ms/step Epoch 10/30 121/121 - 4s - loss: 0.7662 - accuracy: 0.7138 - val_loss: 1.3635 - val_accuracy: 0.6168 - 4s/epoch - 35ms/step Epoch 11/30 121/121 - 4s - loss: 0.6972 - accuracy: 0.7302 - val_loss: 1.3949 - val_accuracy: 0.6238 - 4s/epoch - 32ms/step Epoch 12/30 121/121 - 4s - loss: 0.6870 - accuracy: 0.7450 - val_loss: 1.3392 - val_accuracy: 0.6565 - 4s/epoch - 32ms/step Epoch 13/30 121/121 - 4s - loss: 0.6067 - accuracy: 0.7713 - val_loss: 1.5277 - val_accuracy: 0.5654 - 4s/epoch - 32ms/step Epoch 14/30 121/121 - 4s - loss: 0.5640 - accuracy: 0.7939 - val_loss: 1.6589 - val_accuracy: 0.6028 - 4s/epoch - 34ms/step Epoch 15/30 121/121 - 4s - loss: 0.5522 - accuracy: 0.7946 - val_loss: 1.7809 - val_accuracy: 0.5888 - 4s/epoch - 33ms/step Epoch 16/30 121/121 - 4s - loss: 0.5487 - accuracy: 0.7941 - val_loss: 2.5506 - val_accuracy: 0.4930 - 4s/epoch - 32ms/step Epoch 17/30 121/121 - 4s - loss: 0.5108 - accuracy: 0.8097 - val_loss: 2.0627 - val_accuracy: 0.5047 - 4s/epoch - 34ms/step Epoch 18/30 121/121 - 4s - loss: 0.5019 - accuracy: 0.8149 - val_loss: 1.5702 - val_accuracy: 0.6495 - 4s/epoch - 36ms/step Epoch 19/30 121/121 - 4s - loss: 0.4935 - accuracy: 0.8154 - val_loss: 2.0993 - val_accuracy: 0.6075 - 4s/epoch - 35ms/step Epoch 20/30 121/121 - 4s - loss: 0.4837 - accuracy: 0.8183 - val_loss: 1.7613 - val_accuracy: 0.5888 - 4s/epoch - 32ms/step Epoch 21/30 121/121 - 4s - loss: 0.4792 - accuracy: 0.8290 - val_loss: 1.9179 - val_accuracy: 0.5584 - 4s/epoch - 32ms/step Epoch 22/30 121/121 - 4s - loss: 0.4511 - accuracy: 0.8284 - val_loss: 1.8315 - val_accuracy: 0.5561 - 4s/epoch - 33ms/step Epoch 23/30 121/121 - 4s - loss: 0.4567 - accuracy: 0.8303 - val_loss: 1.7862 - val_accuracy: 0.5654 - 4s/epoch - 35ms/step Epoch 24/30 121/121 - 4s - loss: 0.4266 - accuracy: 0.8373 - val_loss: 2.2484 - val_accuracy: 0.5981 - 4s/epoch - 32ms/step Epoch 25/30 121/121 - 4s - loss: 0.4294 - accuracy: 0.8412 - val_loss: 1.8034 - val_accuracy: 0.6449 - 4s/epoch - 32ms/step Epoch 26/30 121/121 - 4s - loss: 0.4149 - accuracy: 0.8448 - val_loss: 2.0269 - val_accuracy: 0.5981 - 4s/epoch - 33ms/step Epoch 27/30 121/121 - 4s - loss: 0.4074 - accuracy: 0.8498 - val_loss: 1.8727 - val_accuracy: 0.6262 - 4s/epoch - 34ms/step Epoch 28/30 121/121 - 4s - loss: 0.3652 - accuracy: 0.8622 - val_loss: 2.7845 - val_accuracy: 0.5070 - 4s/epoch - 33ms/step Epoch 29/30 121/121 - 4s - loss: 0.3715 - accuracy: 0.8640 - val_loss: 2.0674 - val_accuracy: 0.6098 - 4s/epoch - 33ms/step Epoch 30/30 121/121 - 4s - loss: 0.3674 - accuracy: 0.8653 - val_loss: 2.5580 - val_accuracy: 0.6005 - 4s/epoch - 33ms/step
plt.plot(history2.history['accuracy'])
plt.plot(history2.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model2.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 2.7567 - accuracy: 0.6084 - 495ms/epoch - 33ms/step
y_pred=model2.predict(X_test_normalized)
15/15 [==============================] - 0s 8ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
train_datagen = ImageDataGenerator(
horizontal_flip=True,
vertical_flip=True,
zoom_range=0.75,
rotation_range=20,
fill_mode='nearest'
)
# Intializing a sequential model
model3 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model3.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model3.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model3.add(MaxPooling2D((2, 2), padding = 'same'))
model3.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model3.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model3.add(MaxPooling2D((2, 2), padding = 'same'))
model3.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model3.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model3.add(Dense(16, activation='relu'))
model3.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model3.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model3.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model3.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
conv2d_1 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d (MaxPooling2 (None, 64, 64, 32) 0
D)
conv2d_2 (Conv2D) (None, 64, 64, 32) 9248
conv2d_3 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization (Batch (None, 32, 32, 32) 128
Normalization)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 563388 (2.15 MB)
Trainable params: 563324 (2.15 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
# Epochs
epochs = 30
# Batch size
batch_size = 128
history3 = model3.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
batch_size=batch_size,
seed=42,
shuffle=False),
epochs=epochs,
steps_per_epoch=X_train_normalized.shape[0] // batch_size,
validation_data=(X_val_normalized,y_val_encoded),
verbose=1)
Epoch 1/30 30/30 [==============================] - 24s 589ms/step - loss: 2.3500 - accuracy: 0.2019 - val_loss: 2.4653 - val_accuracy: 0.0841 Epoch 2/30 30/30 [==============================] - 19s 612ms/step - loss: 2.1290 - accuracy: 0.2681 - val_loss: 2.4346 - val_accuracy: 0.1776 Epoch 3/30 30/30 [==============================] - 18s 612ms/step - loss: 2.0140 - accuracy: 0.3084 - val_loss: 2.4100 - val_accuracy: 0.2009 Epoch 4/30 30/30 [==============================] - 18s 609ms/step - loss: 1.9259 - accuracy: 0.3410 - val_loss: 2.3832 - val_accuracy: 0.1729 Epoch 5/30 30/30 [==============================] - 18s 608ms/step - loss: 1.8656 - accuracy: 0.3614 - val_loss: 2.3892 - val_accuracy: 0.2827 Epoch 6/30 30/30 [==============================] - 18s 611ms/step - loss: 1.7888 - accuracy: 0.4012 - val_loss: 2.3216 - val_accuracy: 0.2407 Epoch 7/30 30/30 [==============================] - 18s 604ms/step - loss: 1.7790 - accuracy: 0.3985 - val_loss: 2.3193 - val_accuracy: 0.1799 Epoch 8/30 30/30 [==============================] - 19s 614ms/step - loss: 1.7119 - accuracy: 0.4230 - val_loss: 2.3610 - val_accuracy: 0.1262 Epoch 9/30 30/30 [==============================] - 22s 714ms/step - loss: 1.6444 - accuracy: 0.4396 - val_loss: 2.2978 - val_accuracy: 0.1425 Epoch 10/30 30/30 [==============================] - 18s 610ms/step - loss: 1.6207 - accuracy: 0.4439 - val_loss: 2.1899 - val_accuracy: 0.2336 Epoch 11/30 30/30 [==============================] - 18s 617ms/step - loss: 1.5668 - accuracy: 0.4585 - val_loss: 2.3125 - val_accuracy: 0.2313 Epoch 12/30 30/30 [==============================] - 18s 607ms/step - loss: 1.5623 - accuracy: 0.4665 - val_loss: 2.1595 - val_accuracy: 0.2407 Epoch 13/30 30/30 [==============================] - 30s 998ms/step - loss: 1.5141 - accuracy: 0.4732 - val_loss: 2.0366 - val_accuracy: 0.3294 Epoch 14/30 30/30 [==============================] - 20s 646ms/step - loss: 1.4639 - accuracy: 0.4977 - val_loss: 1.9530 - val_accuracy: 0.4322 Epoch 15/30 30/30 [==============================] - 18s 604ms/step - loss: 1.4814 - accuracy: 0.4945 - val_loss: 1.9067 - val_accuracy: 0.4019 Epoch 16/30 30/30 [==============================] - 18s 617ms/step - loss: 1.4457 - accuracy: 0.5069 - val_loss: 1.8728 - val_accuracy: 0.3505 Epoch 17/30 30/30 [==============================] - 18s 613ms/step - loss: 1.4740 - accuracy: 0.4870 - val_loss: 1.6365 - val_accuracy: 0.5187 Epoch 18/30 30/30 [==============================] - 19s 627ms/step - loss: 1.4379 - accuracy: 0.4993 - val_loss: 1.6905 - val_accuracy: 0.4159 Epoch 19/30 30/30 [==============================] - 22s 707ms/step - loss: 1.4318 - accuracy: 0.5069 - val_loss: 1.7418 - val_accuracy: 0.4019 Epoch 20/30 30/30 [==============================] - 18s 613ms/step - loss: 1.3875 - accuracy: 0.5233 - val_loss: 1.5510 - val_accuracy: 0.4136 Epoch 21/30 30/30 [==============================] - 19s 619ms/step - loss: 1.3157 - accuracy: 0.5450 - val_loss: 1.9297 - val_accuracy: 0.3341 Epoch 22/30 30/30 [==============================] - 19s 629ms/step - loss: 1.3154 - accuracy: 0.5485 - val_loss: 1.6098 - val_accuracy: 0.4603 Epoch 23/30 30/30 [==============================] - 20s 636ms/step - loss: 1.3179 - accuracy: 0.5531 - val_loss: 2.1655 - val_accuracy: 0.3178 Epoch 24/30 30/30 [==============================] - 19s 623ms/step - loss: 1.3448 - accuracy: 0.5445 - val_loss: 4.8545 - val_accuracy: 0.1238 Epoch 25/30 30/30 [==============================] - 18s 615ms/step - loss: 1.3991 - accuracy: 0.5305 - val_loss: 2.3973 - val_accuracy: 0.3528 Epoch 26/30 30/30 [==============================] - 19s 620ms/step - loss: 1.2736 - accuracy: 0.5579 - val_loss: 1.1037 - val_accuracy: 0.6168 Epoch 27/30 30/30 [==============================] - 19s 622ms/step - loss: 1.2642 - accuracy: 0.5668 - val_loss: 2.1385 - val_accuracy: 0.2734 Epoch 28/30 30/30 [==============================] - 19s 633ms/step - loss: 1.2577 - accuracy: 0.5700 - val_loss: 1.1402 - val_accuracy: 0.6285 Epoch 29/30 30/30 [==============================] - 19s 623ms/step - loss: 1.2437 - accuracy: 0.5840 - val_loss: 1.1927 - val_accuracy: 0.6168 Epoch 30/30 30/30 [==============================] - 19s 627ms/step - loss: 1.2182 - accuracy: 0.5859 - val_loss: 2.0179 - val_accuracy: 0.4533
plt.plot(history3.history['accuracy'])
plt.plot(history3.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model3.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 2.0331 - accuracy: 0.4632 - 263ms/epoch - 18ms/step
y_pred=model3.predict(X_test_normalized)
15/15 [==============================] - 0s 12ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model4 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model4.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model4.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model4.add(MaxPooling2D((2, 2), padding = 'same'))
model4.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model4.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model4.add(MaxPooling2D((2, 2), padding = 'same'))
model4.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model4.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model4.add(Dense(16, activation='relu'))
model4.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model4.add(Dense(12, activation='softmax'))
#using SDG optimizer
opt = SGD(learning_rate=0.01, momentum=0.9)
# Compile model
model4.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model4.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
conv2d_1 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d (MaxPooling2 (None, 64, 64, 32) 0
D)
conv2d_2 (Conv2D) (None, 64, 64, 32) 9248
conv2d_3 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization (Batch (None, 32, 32, 32) 128
Normalization)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 563388 (2.15 MB)
Trainable params: 563324 (2.15 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
# Epochs
epochs = 30
# Batch size
batch_size = 128
history4 = model4.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
batch_size=batch_size,
seed=42,
shuffle=False),
epochs=epochs,
steps_per_epoch=X_train_normalized.shape[0] // batch_size,
validation_data=(X_val_normalized,y_val_encoded),
verbose=1)
Epoch 1/30 30/30 [==============================] - 21s 643ms/step - loss: 2.4236 - accuracy: 0.1592 - val_loss: 2.7105 - val_accuracy: 0.1379 Epoch 2/30 30/30 [==============================] - 20s 666ms/step - loss: 2.2585 - accuracy: 0.2277 - val_loss: 2.9497 - val_accuracy: 0.1379 Epoch 3/30 30/30 [==============================] - 18s 606ms/step - loss: 2.1569 - accuracy: 0.2603 - val_loss: 3.4194 - val_accuracy: 0.1379 Epoch 4/30 30/30 [==============================] - 18s 614ms/step - loss: 2.1469 - accuracy: 0.2358 - val_loss: 4.3010 - val_accuracy: 0.1379 Epoch 5/30 30/30 [==============================] - 19s 622ms/step - loss: 2.0584 - accuracy: 0.2780 - val_loss: 2.3405 - val_accuracy: 0.2173 Epoch 6/30 30/30 [==============================] - 19s 631ms/step - loss: 1.9832 - accuracy: 0.2939 - val_loss: 3.1069 - val_accuracy: 0.1379 Epoch 7/30 30/30 [==============================] - 18s 614ms/step - loss: 1.9861 - accuracy: 0.2971 - val_loss: 2.4476 - val_accuracy: 0.1519 Epoch 8/30 30/30 [==============================] - 18s 600ms/step - loss: 1.9600 - accuracy: 0.3087 - val_loss: 2.4973 - val_accuracy: 0.2150 Epoch 9/30 30/30 [==============================] - 18s 615ms/step - loss: 1.9309 - accuracy: 0.3065 - val_loss: 9.0049 - val_accuracy: 0.1379 Epoch 10/30 30/30 [==============================] - 18s 612ms/step - loss: 1.9303 - accuracy: 0.3151 - val_loss: 4.5997 - val_accuracy: 0.0607 Epoch 11/30 30/30 [==============================] - 18s 607ms/step - loss: 1.9094 - accuracy: 0.3200 - val_loss: 4.7515 - val_accuracy: 0.1542 Epoch 12/30 30/30 [==============================] - 19s 605ms/step - loss: 1.8973 - accuracy: 0.3307 - val_loss: 2.5933 - val_accuracy: 0.2640 Epoch 13/30 30/30 [==============================] - 19s 618ms/step - loss: 1.8733 - accuracy: 0.3358 - val_loss: 11.4476 - val_accuracy: 0.0514 Epoch 14/30 30/30 [==============================] - 21s 699ms/step - loss: 1.8551 - accuracy: 0.3463 - val_loss: 7.9070 - val_accuracy: 0.0607 Epoch 15/30 30/30 [==============================] - 19s 617ms/step - loss: 1.8658 - accuracy: 0.3482 - val_loss: 2.5382 - val_accuracy: 0.3224 Epoch 16/30 30/30 [==============================] - 18s 603ms/step - loss: 1.9032 - accuracy: 0.3345 - val_loss: 6.5856 - val_accuracy: 0.1472 Epoch 17/30 30/30 [==============================] - 18s 600ms/step - loss: 1.8351 - accuracy: 0.3595 - val_loss: 20.6634 - val_accuracy: 0.1425 Epoch 18/30 30/30 [==============================] - 18s 605ms/step - loss: 1.8218 - accuracy: 0.3625 - val_loss: 4.9153 - val_accuracy: 0.1893 Epoch 19/30 30/30 [==============================] - 18s 601ms/step - loss: 1.7856 - accuracy: 0.3845 - val_loss: 15.0511 - val_accuracy: 0.1262 Epoch 20/30 30/30 [==============================] - 18s 599ms/step - loss: 1.7389 - accuracy: 0.4028 - val_loss: 7.3397 - val_accuracy: 0.1238 Epoch 21/30 30/30 [==============================] - 19s 624ms/step - loss: 1.7341 - accuracy: 0.4023 - val_loss: 8.3138 - val_accuracy: 0.1168 Epoch 22/30 30/30 [==============================] - 19s 627ms/step - loss: 1.7279 - accuracy: 0.4106 - val_loss: 3.9766 - val_accuracy: 0.1846 Epoch 23/30 30/30 [==============================] - 18s 603ms/step - loss: 1.6976 - accuracy: 0.4157 - val_loss: 18.9211 - val_accuracy: 0.1168 Epoch 24/30 30/30 [==============================] - 18s 614ms/step - loss: 1.7635 - accuracy: 0.4071 - val_loss: 16.1108 - val_accuracy: 0.1028 Epoch 25/30 30/30 [==============================] - 18s 603ms/step - loss: 2.1114 - accuracy: 0.2616 - val_loss: 31.7079 - val_accuracy: 0.0537 Epoch 26/30 30/30 [==============================] - 19s 633ms/step - loss: 1.7839 - accuracy: 0.3673 - val_loss: 15.2381 - val_accuracy: 0.0818 Epoch 27/30 30/30 [==============================] - 20s 664ms/step - loss: 1.6943 - accuracy: 0.3958 - val_loss: 26.4368 - val_accuracy: 0.1075 Epoch 28/30 30/30 [==============================] - 19s 616ms/step - loss: 1.6917 - accuracy: 0.4119 - val_loss: 13.3201 - val_accuracy: 0.1402 Epoch 29/30 30/30 [==============================] - 19s 607ms/step - loss: 1.7584 - accuracy: 0.4006 - val_loss: 19.2734 - val_accuracy: 0.0888 Epoch 30/30 30/30 [==============================] - 19s 619ms/step - loss: 1.6518 - accuracy: 0.4195 - val_loss: 5.6886 - val_accuracy: 0.2383
plt.plot(history4.history['accuracy'])
plt.plot(history4.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model4.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 5.7588 - accuracy: 0.2379 - 212ms/epoch - 14ms/step
y_pred=model4.predict(X_test_normalized)
15/15 [==============================] - 0s 13ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model5 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model5.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
# Adding max pooling to reduce the size of output of first conv layer
model5.add(MaxPooling2D((2, 2), padding = 'same'))
model5.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model5.add(MaxPooling2D((2, 2), padding = 'same'))
model5.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model5.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model5.add(Dense(16, activation='relu'))
model5.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model5.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt = Adam()
# Compile model
model5.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model5.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
max_pooling2d (MaxPooling2 (None, 64, 64, 64) 0
D)
conv2d_1 (Conv2D) (None, 64, 64, 32) 18464
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization (Batch (None, 32, 32, 32) 128
Normalization)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 544892 (2.08 MB)
Trainable params: 544828 (2.08 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
# Epochs
epochs = 30
# Batch size
batch_size = 128
history5 = model5.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
batch_size=batch_size,
seed=42,
shuffle=False),
epochs=epochs,
steps_per_epoch=X_train_normalized.shape[0] // batch_size,
validation_data=(X_val_normalized,y_val_encoded),
verbose=1)
Epoch 1/30 30/30 [==============================] - 20s 571ms/step - loss: 2.3523 - accuracy: 0.1788 - val_loss: 2.4396 - val_accuracy: 0.1869 Epoch 2/30 30/30 [==============================] - 18s 587ms/step - loss: 2.0984 - accuracy: 0.2592 - val_loss: 2.3821 - val_accuracy: 0.2617 Epoch 3/30 30/30 [==============================] - 18s 579ms/step - loss: 2.0206 - accuracy: 0.2955 - val_loss: 2.3669 - val_accuracy: 0.1612 Epoch 4/30 30/30 [==============================] - 18s 590ms/step - loss: 1.9505 - accuracy: 0.3275 - val_loss: 2.3517 - val_accuracy: 0.3458 Epoch 5/30 30/30 [==============================] - 18s 587ms/step - loss: 1.8589 - accuracy: 0.3469 - val_loss: 2.2980 - val_accuracy: 0.2664 Epoch 6/30 30/30 [==============================] - 18s 591ms/step - loss: 1.8399 - accuracy: 0.3617 - val_loss: 2.2984 - val_accuracy: 0.3505 Epoch 7/30 30/30 [==============================] - 18s 588ms/step - loss: 1.8230 - accuracy: 0.3670 - val_loss: 2.3469 - val_accuracy: 0.2243 Epoch 8/30 30/30 [==============================] - 17s 581ms/step - loss: 1.7633 - accuracy: 0.3786 - val_loss: 2.2897 - val_accuracy: 0.3879 Epoch 9/30 30/30 [==============================] - 21s 709ms/step - loss: 1.7460 - accuracy: 0.3947 - val_loss: 2.2246 - val_accuracy: 0.2827 Epoch 10/30 30/30 [==============================] - 18s 591ms/step - loss: 1.7286 - accuracy: 0.3928 - val_loss: 2.2293 - val_accuracy: 0.3715 Epoch 11/30 30/30 [==============================] - 18s 592ms/step - loss: 1.7333 - accuracy: 0.4025 - val_loss: 2.2314 - val_accuracy: 0.3715 Epoch 12/30 30/30 [==============================] - 18s 600ms/step - loss: 1.6694 - accuracy: 0.4122 - val_loss: 2.1726 - val_accuracy: 0.4579 Epoch 13/30 30/30 [==============================] - 18s 587ms/step - loss: 1.6293 - accuracy: 0.4243 - val_loss: 1.8488 - val_accuracy: 0.3972 Epoch 14/30 30/30 [==============================] - 18s 586ms/step - loss: 1.6726 - accuracy: 0.4138 - val_loss: 1.9667 - val_accuracy: 0.4813 Epoch 15/30 30/30 [==============================] - 18s 589ms/step - loss: 1.6103 - accuracy: 0.4316 - val_loss: 2.2121 - val_accuracy: 0.2430 Epoch 16/30 30/30 [==============================] - 18s 591ms/step - loss: 1.6505 - accuracy: 0.4297 - val_loss: 1.8490 - val_accuracy: 0.4486 Epoch 17/30 30/30 [==============================] - 18s 595ms/step - loss: 1.6338 - accuracy: 0.4343 - val_loss: 1.7635 - val_accuracy: 0.4930 Epoch 18/30 30/30 [==============================] - 18s 586ms/step - loss: 1.5991 - accuracy: 0.4367 - val_loss: 1.5619 - val_accuracy: 0.5047 Epoch 19/30 30/30 [==============================] - 18s 591ms/step - loss: 1.5965 - accuracy: 0.4431 - val_loss: 1.8394 - val_accuracy: 0.5117 Epoch 20/30 30/30 [==============================] - 18s 593ms/step - loss: 1.5267 - accuracy: 0.4668 - val_loss: 1.4332 - val_accuracy: 0.4977 Epoch 21/30 30/30 [==============================] - 21s 707ms/step - loss: 1.4737 - accuracy: 0.4781 - val_loss: 1.5091 - val_accuracy: 0.4930 Epoch 22/30 30/30 [==============================] - 18s 595ms/step - loss: 1.5239 - accuracy: 0.4727 - val_loss: 1.4079 - val_accuracy: 0.5818 Epoch 23/30 30/30 [==============================] - 23s 763ms/step - loss: 1.5225 - accuracy: 0.4784 - val_loss: 1.3036 - val_accuracy: 0.5491 Epoch 24/30 30/30 [==============================] - 18s 583ms/step - loss: 1.4657 - accuracy: 0.4886 - val_loss: 1.5022 - val_accuracy: 0.5070 Epoch 25/30 30/30 [==============================] - 18s 584ms/step - loss: 1.4812 - accuracy: 0.4775 - val_loss: 1.2691 - val_accuracy: 0.5794 Epoch 26/30 30/30 [==============================] - 18s 594ms/step - loss: 1.4138 - accuracy: 0.5004 - val_loss: 1.1128 - val_accuracy: 0.6519 Epoch 27/30 30/30 [==============================] - 18s 598ms/step - loss: 1.4621 - accuracy: 0.4824 - val_loss: 1.1321 - val_accuracy: 0.6098 Epoch 28/30 30/30 [==============================] - 19s 606ms/step - loss: 1.4314 - accuracy: 0.5044 - val_loss: 1.1114 - val_accuracy: 0.6729 Epoch 29/30 30/30 [==============================] - 18s 586ms/step - loss: 1.4020 - accuracy: 0.5020 - val_loss: 1.4814 - val_accuracy: 0.5304 Epoch 30/30 30/30 [==============================] - 18s 597ms/step - loss: 1.4148 - accuracy: 0.5052 - val_loss: 1.5126 - val_accuracy: 0.4650
plt.plot(history5.history['accuracy'])
plt.plot(history5.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model5.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 1.6379 - accuracy: 0.4400 - 134ms/epoch - 9ms/step
y_pred=model5.predict(X_test_normalized)
15/15 [==============================] - 0s 7ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
from keras.applications.vgg16 import VGG16
vgg_model1 = VGG16(weights='imagenet', include_top = False, input_shape = (128,128,3))
vgg_model1.summary()
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58889256/58889256 [==============================] - 2s 0us/step
Model: "vgg16"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 128, 3)] 0
block1_conv1 (Conv2D) (None, 128, 128, 64) 1792
block1_conv2 (Conv2D) (None, 128, 128, 64) 36928
block1_pool (MaxPooling2D) (None, 64, 64, 64) 0
block2_conv1 (Conv2D) (None, 64, 64, 128) 73856
block2_conv2 (Conv2D) (None, 64, 64, 128) 147584
block2_pool (MaxPooling2D) (None, 32, 32, 128) 0
block3_conv1 (Conv2D) (None, 32, 32, 256) 295168
block3_conv2 (Conv2D) (None, 32, 32, 256) 590080
block3_conv3 (Conv2D) (None, 32, 32, 256) 590080
block3_pool (MaxPooling2D) (None, 16, 16, 256) 0
block4_conv1 (Conv2D) (None, 16, 16, 512) 1180160
block4_conv2 (Conv2D) (None, 16, 16, 512) 2359808
block4_conv3 (Conv2D) (None, 16, 16, 512) 2359808
block4_pool (MaxPooling2D) (None, 8, 8, 512) 0
block5_conv1 (Conv2D) (None, 8, 8, 512) 2359808
block5_conv2 (Conv2D) (None, 8, 8, 512) 2359808
block5_conv3 (Conv2D) (None, 8, 8, 512) 2359808
block5_pool (MaxPooling2D) (None, 4, 4, 512) 0
=================================================================
Total params: 14714688 (56.13 MB)
Trainable params: 14714688 (56.13 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
# Making all the layers of the VGG model non-trainable. i.e. freezing them
for layer in vgg_model1.layers:
layer.trainable = False
model6 = Sequential()
# Adding the convolutional part of the VGG16 model from above
model6.add(vgg_model1)
# Flattening the output of the VGG16 model because it is from a convolutional layer
model6.add(Flatten())
# Adding a fully connected dense layer
model6.add(Dense(16, activation='relu'))
model6.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model6.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt = Adam()
# Compile model
model6.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model6.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
vgg16 (Functional) (None, 4, 4, 512) 14714688
flatten_1 (Flatten) (None, 8192) 0
dense_2 (Dense) (None, 16) 131088
dropout_1 (Dropout) (None, 16) 0
dense_3 (Dense) (None, 12) 204
=================================================================
Total params: 14845980 (56.63 MB)
Trainable params: 131292 (512.86 KB)
Non-trainable params: 14714688 (56.13 MB)
_________________________________________________________________
# Epochs
epochs = 30
# Batch size
batch_size = 128
history6 = model6.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
batch_size=batch_size,
seed=42,
shuffle=False),
epochs=epochs,
steps_per_epoch=X_train_normalized.shape[0] // batch_size,
validation_data=(X_val_normalized,y_val_encoded),
verbose=1)
Epoch 1/30 30/30 [==============================] - 31s 825ms/step - loss: 2.4394 - accuracy: 0.1433 - val_loss: 2.3099 - val_accuracy: 0.1682 Epoch 2/30 30/30 [==============================] - 20s 658ms/step - loss: 2.3494 - accuracy: 0.1662 - val_loss: 2.2923 - val_accuracy: 0.2173 Epoch 3/30 30/30 [==============================] - 19s 643ms/step - loss: 2.2901 - accuracy: 0.1845 - val_loss: 2.2151 - val_accuracy: 0.2687 Epoch 4/30 30/30 [==============================] - 19s 641ms/step - loss: 2.2438 - accuracy: 0.2208 - val_loss: 2.1504 - val_accuracy: 0.2874 Epoch 5/30 30/30 [==============================] - 20s 646ms/step - loss: 2.2162 - accuracy: 0.2264 - val_loss: 2.0888 - val_accuracy: 0.2874 Epoch 6/30 30/30 [==============================] - 20s 651ms/step - loss: 2.1796 - accuracy: 0.2194 - val_loss: 2.0147 - val_accuracy: 0.2780 Epoch 7/30 30/30 [==============================] - 19s 636ms/step - loss: 2.1476 - accuracy: 0.2393 - val_loss: 1.9719 - val_accuracy: 0.2944 Epoch 8/30 30/30 [==============================] - 20s 661ms/step - loss: 2.1171 - accuracy: 0.2501 - val_loss: 1.9975 - val_accuracy: 0.3481 Epoch 9/30 30/30 [==============================] - 19s 648ms/step - loss: 2.1047 - accuracy: 0.2597 - val_loss: 1.9454 - val_accuracy: 0.3248 Epoch 10/30 30/30 [==============================] - 19s 640ms/step - loss: 2.0974 - accuracy: 0.2544 - val_loss: 1.9068 - val_accuracy: 0.3435 Epoch 11/30 30/30 [==============================] - 20s 657ms/step - loss: 2.0694 - accuracy: 0.2581 - val_loss: 1.9249 - val_accuracy: 0.3668 Epoch 12/30 30/30 [==============================] - 25s 735ms/step - loss: 2.0615 - accuracy: 0.2670 - val_loss: 1.9023 - val_accuracy: 0.3738 Epoch 13/30 30/30 [==============================] - 19s 648ms/step - loss: 2.0362 - accuracy: 0.2681 - val_loss: 1.8797 - val_accuracy: 0.3575 Epoch 14/30 30/30 [==============================] - 25s 823ms/step - loss: 2.0460 - accuracy: 0.2716 - val_loss: 1.8429 - val_accuracy: 0.3762 Epoch 15/30 30/30 [==============================] - 35s 1s/step - loss: 2.0234 - accuracy: 0.2695 - val_loss: 1.8411 - val_accuracy: 0.3925 Epoch 16/30 30/30 [==============================] - 30s 983ms/step - loss: 2.0107 - accuracy: 0.2740 - val_loss: 1.8043 - val_accuracy: 0.3902 Epoch 17/30 30/30 [==============================] - 23s 728ms/step - loss: 2.0089 - accuracy: 0.2713 - val_loss: 1.8451 - val_accuracy: 0.3832 Epoch 18/30 30/30 [==============================] - 21s 702ms/step - loss: 1.9935 - accuracy: 0.2831 - val_loss: 1.8290 - val_accuracy: 0.3879 Epoch 19/30 30/30 [==============================] - 20s 647ms/step - loss: 1.9770 - accuracy: 0.2710 - val_loss: 1.8091 - val_accuracy: 0.4136 Epoch 20/30 30/30 [==============================] - 20s 651ms/step - loss: 1.9592 - accuracy: 0.2753 - val_loss: 1.7678 - val_accuracy: 0.4206 Epoch 21/30 30/30 [==============================] - 20s 664ms/step - loss: 1.9694 - accuracy: 0.2853 - val_loss: 1.7406 - val_accuracy: 0.4089 Epoch 22/30 30/30 [==============================] - 23s 788ms/step - loss: 1.9622 - accuracy: 0.2791 - val_loss: 1.7844 - val_accuracy: 0.4393 Epoch 23/30 30/30 [==============================] - 20s 668ms/step - loss: 1.9702 - accuracy: 0.2829 - val_loss: 1.7191 - val_accuracy: 0.4276 Epoch 24/30 30/30 [==============================] - 19s 665ms/step - loss: 1.9543 - accuracy: 0.2952 - val_loss: 1.7531 - val_accuracy: 0.4089 Epoch 25/30 30/30 [==============================] - 19s 647ms/step - loss: 1.9340 - accuracy: 0.2861 - val_loss: 1.7221 - val_accuracy: 0.4322 Epoch 26/30 30/30 [==============================] - 19s 623ms/step - loss: 1.9236 - accuracy: 0.2947 - val_loss: 1.7315 - val_accuracy: 0.4322 Epoch 27/30 30/30 [==============================] - 21s 673ms/step - loss: 1.9320 - accuracy: 0.2899 - val_loss: 1.7336 - val_accuracy: 0.4346 Epoch 28/30 30/30 [==============================] - 20s 655ms/step - loss: 1.9459 - accuracy: 0.2796 - val_loss: 1.7181 - val_accuracy: 0.4416 Epoch 29/30 30/30 [==============================] - 19s 629ms/step - loss: 1.9407 - accuracy: 0.2861 - val_loss: 1.7504 - val_accuracy: 0.4276 Epoch 30/30 30/30 [==============================] - 19s 633ms/step - loss: 1.9317 - accuracy: 0.2845 - val_loss: 1.6842 - val_accuracy: 0.4346
plt.plot(history6.history['accuracy'])
plt.plot(history6.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model6.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 1s - loss: 1.6832 - accuracy: 0.4358 - 705ms/epoch - 47ms/step
y_pred=model6.predict(X_test_normalized)
15/15 [==============================] - 1s 42ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
vgg_model2 = Model(inputs=vgg_model1.inputs,outputs=vgg_model1.layers[6].output)
vgg_model2.summary()
Model: "model_3"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 128, 3)] 0
block1_conv1 (Conv2D) (None, 128, 128, 64) 1792
block1_conv2 (Conv2D) (None, 128, 128, 64) 36928
block1_pool (MaxPooling2D) (None, 64, 64, 64) 0
block2_conv1 (Conv2D) (None, 64, 64, 128) 73856
block2_conv2 (Conv2D) (None, 64, 64, 128) 147584
block2_pool (MaxPooling2D) (None, 32, 32, 128) 0
=================================================================
Total params: 260160 (1016.25 KB)
Trainable params: 0 (0.00 Byte)
Non-trainable params: 260160 (1016.25 KB)
_________________________________________________________________
# Making all the layers of the VGG model non-trainable. i.e. freezing them
for layer in vgg_model2.layers:
layer.trainable = False
model7 = Sequential()
# Adding the convolutional part of the VGG16 model from above
model7.add(vgg_model2)
# Flattening the output of the VGG16 model because it is from a convolutional layer
model7.add(Flatten())
# Adding a fully connected dense layer
model7.add(Dense(16, activation='relu'))
model7.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model7.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt = Adam()
# Compile model
model7.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model7.summary()
Model: "sequential_4"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
model_3 (Functional) (None, 32, 32, 128) 260160
flatten_4 (Flatten) (None, 131072) 0
dense_8 (Dense) (None, 16) 2097168
dropout_4 (Dropout) (None, 16) 0
dense_9 (Dense) (None, 12) 204
=================================================================
Total params: 2357532 (8.99 MB)
Trainable params: 2097372 (8.00 MB)
Non-trainable params: 260160 (1016.25 KB)
_________________________________________________________________
# Epochs
epochs = 30
# Batch size
batch_size = 128
history7 = model7.fit(train_datagen.flow(X_train_normalized,y_train_encoded,
batch_size=batch_size,
seed=42,
shuffle=False),
epochs=epochs,
steps_per_epoch=X_train_normalized.shape[0] // batch_size,
validation_data=(X_val_normalized,y_val_encoded),
verbose=1)
Epoch 1/30 30/30 [==============================] - 24s 782ms/step - loss: 27.2218 - accuracy: 0.1199 - val_loss: 2.4812 - val_accuracy: 0.1285 Epoch 2/30 30/30 [==============================] - 21s 702ms/step - loss: 2.4789 - accuracy: 0.1288 - val_loss: 2.4767 - val_accuracy: 0.1285 Epoch 3/30 30/30 [==============================] - 21s 676ms/step - loss: 2.4744 - accuracy: 0.1293 - val_loss: 2.4720 - val_accuracy: 0.1285 Epoch 4/30 30/30 [==============================] - 18s 615ms/step - loss: 2.4696 - accuracy: 0.1288 - val_loss: 2.4674 - val_accuracy: 0.1285 Epoch 5/30 30/30 [==============================] - 19s 625ms/step - loss: 2.4652 - accuracy: 0.1284 - val_loss: 2.4631 - val_accuracy: 0.1285 Epoch 6/30 30/30 [==============================] - 22s 733ms/step - loss: 2.4612 - accuracy: 0.1277 - val_loss: 2.4593 - val_accuracy: 0.1285 Epoch 7/30 30/30 [==============================] - 19s 609ms/step - loss: 2.4574 - accuracy: 0.1272 - val_loss: 2.4555 - val_accuracy: 0.1285 Epoch 8/30 30/30 [==============================] - 19s 637ms/step - loss: 2.4534 - accuracy: 0.1285 - val_loss: 2.4520 - val_accuracy: 0.1285 Epoch 9/30 30/30 [==============================] - 18s 609ms/step - loss: 2.4497 - accuracy: 0.1288 - val_loss: 2.4486 - val_accuracy: 0.1285 Epoch 10/30 30/30 [==============================] - 18s 609ms/step - loss: 2.4470 - accuracy: 0.1272 - val_loss: 2.4458 - val_accuracy: 0.1285 Epoch 11/30 30/30 [==============================] - 18s 601ms/step - loss: 2.4440 - accuracy: 0.1285 - val_loss: 2.4431 - val_accuracy: 0.1285 Epoch 12/30 30/30 [==============================] - 18s 593ms/step - loss: 2.4416 - accuracy: 0.1272 - val_loss: 2.4406 - val_accuracy: 0.1285 Epoch 13/30 30/30 [==============================] - 19s 621ms/step - loss: 2.4384 - accuracy: 0.1288 - val_loss: 2.4384 - val_accuracy: 0.1285 Epoch 14/30 30/30 [==============================] - 19s 649ms/step - loss: 2.4365 - accuracy: 0.1293 - val_loss: 2.4364 - val_accuracy: 0.1285 Epoch 15/30 30/30 [==============================] - 18s 632ms/step - loss: 2.4351 - accuracy: 0.1261 - val_loss: 2.4345 - val_accuracy: 0.1285 Epoch 16/30 30/30 [==============================] - 19s 613ms/step - loss: 2.4317 - accuracy: 0.1288 - val_loss: 2.4328 - val_accuracy: 0.1285 Epoch 17/30 30/30 [==============================] - 18s 608ms/step - loss: 2.4315 - accuracy: 0.1285 - val_loss: 2.4312 - val_accuracy: 0.1285 Epoch 18/30 30/30 [==============================] - 18s 604ms/step - loss: 2.4301 - accuracy: 0.1261 - val_loss: 2.4299 - val_accuracy: 0.1285 Epoch 19/30 30/30 [==============================] - 18s 614ms/step - loss: 2.4270 - accuracy: 0.1301 - val_loss: 2.4285 - val_accuracy: 0.1285 Epoch 20/30 30/30 [==============================] - 18s 603ms/step - loss: 2.4257 - accuracy: 0.1301 - val_loss: 2.4273 - val_accuracy: 0.1285 Epoch 21/30 30/30 [==============================] - 18s 595ms/step - loss: 2.4245 - accuracy: 0.1301 - val_loss: 2.4262 - val_accuracy: 0.1285 Epoch 22/30 30/30 [==============================] - 18s 612ms/step - loss: 2.4244 - accuracy: 0.1293 - val_loss: 2.4254 - val_accuracy: 0.1285 Epoch 23/30 30/30 [==============================] - 19s 612ms/step - loss: 2.4252 - accuracy: 0.1293 - val_loss: 2.4247 - val_accuracy: 0.1285 Epoch 24/30 30/30 [==============================] - 19s 623ms/step - loss: 2.4217 - accuracy: 0.1288 - val_loss: 2.4239 - val_accuracy: 0.1379 Epoch 25/30 30/30 [==============================] - 18s 605ms/step - loss: 2.4226 - accuracy: 0.1390 - val_loss: 2.4232 - val_accuracy: 0.1379 Epoch 26/30 30/30 [==============================] - 22s 732ms/step - loss: 2.4228 - accuracy: 0.1385 - val_loss: 2.4226 - val_accuracy: 0.1379 Epoch 27/30 30/30 [==============================] - 18s 635ms/step - loss: 2.4205 - accuracy: 0.1390 - val_loss: 2.4221 - val_accuracy: 0.1379 Epoch 28/30 30/30 [==============================] - 18s 609ms/step - loss: 2.4213 - accuracy: 0.1374 - val_loss: 2.4216 - val_accuracy: 0.1379 Epoch 29/30 30/30 [==============================] - 19s 620ms/step - loss: 2.4208 - accuracy: 0.1374 - val_loss: 2.4212 - val_accuracy: 0.1379 Epoch 30/30 30/30 [==============================] - 21s 700ms/step - loss: 2.4210 - accuracy: 0.1361 - val_loss: 2.4208 - val_accuracy: 0.1379
plt.plot(history7.history['accuracy'])
plt.plot(history7.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model7.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 2.4194 - accuracy: 0.1368 - 326ms/epoch - 22ms/step
y_pred=model7.predict(X_test_normalized)
15/15 [==============================] - 0s 19ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model8 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model8.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model8.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model8.add(MaxPooling2D((2, 2), padding = 'same'))
model8.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model8.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model8.add(MaxPooling2D((2, 2), padding = 'same'))
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model8.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model8.add(Dense(16, activation='relu'))
model8.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model8.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model8.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model8.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
conv2d_1 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d (MaxPooling2 (None, 64, 64, 32) 0
D)
conv2d_2 (Conv2D) (None, 64, 64, 32) 9248
conv2d_3 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 563260 (2.15 MB)
Trainable params: 563260 (2.15 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
history8 = model8.fit(
X_train_normalized, y_train_encoded,
epochs=30,
validation_data=(X_val_normalized,y_val_encoded),
batch_size=32,
verbose=2
)
Epoch 1/30 121/121 - 6s - loss: 2.4479 - accuracy: 0.1131 - val_loss: 2.4349 - val_accuracy: 0.1168 - 6s/epoch - 53ms/step Epoch 2/30 121/121 - 4s - loss: 2.2894 - accuracy: 0.2371 - val_loss: 2.1182 - val_accuracy: 0.3201 - 4s/epoch - 34ms/step Epoch 3/30 121/121 - 4s - loss: 2.1361 - accuracy: 0.2802 - val_loss: 2.0835 - val_accuracy: 0.3481 - 4s/epoch - 35ms/step Epoch 4/30 121/121 - 4s - loss: 2.0529 - accuracy: 0.2969 - val_loss: 1.9439 - val_accuracy: 0.3575 - 4s/epoch - 34ms/step Epoch 5/30 121/121 - 4s - loss: 2.0064 - accuracy: 0.3112 - val_loss: 2.0102 - val_accuracy: 0.3248 - 4s/epoch - 34ms/step Epoch 6/30 121/121 - 4s - loss: 1.9876 - accuracy: 0.3065 - val_loss: 1.8199 - val_accuracy: 0.3808 - 4s/epoch - 33ms/step Epoch 7/30 121/121 - 4s - loss: 1.9569 - accuracy: 0.3236 - val_loss: 1.7626 - val_accuracy: 0.4159 - 4s/epoch - 32ms/step Epoch 8/30 121/121 - 4s - loss: 1.9218 - accuracy: 0.3314 - val_loss: 1.8143 - val_accuracy: 0.3621 - 4s/epoch - 34ms/step Epoch 9/30 121/121 - 4s - loss: 1.8911 - accuracy: 0.3343 - val_loss: 1.7515 - val_accuracy: 0.4136 - 4s/epoch - 35ms/step Epoch 10/30 121/121 - 4s - loss: 1.8526 - accuracy: 0.3488 - val_loss: 1.7654 - val_accuracy: 0.3692 - 4s/epoch - 33ms/step Epoch 11/30 121/121 - 4s - loss: 1.8520 - accuracy: 0.3322 - val_loss: 1.6924 - val_accuracy: 0.3808 - 4s/epoch - 32ms/step Epoch 12/30 121/121 - 4s - loss: 1.8454 - accuracy: 0.3353 - val_loss: 1.6804 - val_accuracy: 0.4159 - 4s/epoch - 33ms/step Epoch 13/30 121/121 - 4s - loss: 1.8221 - accuracy: 0.3426 - val_loss: 1.7031 - val_accuracy: 0.3808 - 4s/epoch - 34ms/step Epoch 14/30 121/121 - 4s - loss: 1.7809 - accuracy: 0.3533 - val_loss: 1.7013 - val_accuracy: 0.3995 - 4s/epoch - 32ms/step Epoch 15/30 121/121 - 4s - loss: 1.7747 - accuracy: 0.3517 - val_loss: 1.6662 - val_accuracy: 0.4159 - 4s/epoch - 32ms/step Epoch 16/30 121/121 - 4s - loss: 1.7691 - accuracy: 0.3494 - val_loss: 1.6319 - val_accuracy: 0.4159 - 4s/epoch - 33ms/step Epoch 17/30 121/121 - 4s - loss: 1.7410 - accuracy: 0.3613 - val_loss: 1.6213 - val_accuracy: 0.4136 - 4s/epoch - 34ms/step Epoch 18/30 121/121 - 4s - loss: 1.7357 - accuracy: 0.3665 - val_loss: 1.6667 - val_accuracy: 0.4136 - 4s/epoch - 33ms/step Epoch 19/30 121/121 - 4s - loss: 1.7210 - accuracy: 0.3637 - val_loss: 1.6320 - val_accuracy: 0.4182 - 4s/epoch - 32ms/step Epoch 20/30 121/121 - 4s - loss: 1.7157 - accuracy: 0.3696 - val_loss: 1.6198 - val_accuracy: 0.4229 - 4s/epoch - 33ms/step Epoch 21/30 121/121 - 4s - loss: 1.7162 - accuracy: 0.3689 - val_loss: 1.6287 - val_accuracy: 0.4299 - 4s/epoch - 35ms/step Epoch 22/30 121/121 - 4s - loss: 1.7219 - accuracy: 0.3600 - val_loss: 1.6148 - val_accuracy: 0.4136 - 4s/epoch - 33ms/step Epoch 23/30 121/121 - 4s - loss: 1.6884 - accuracy: 0.3754 - val_loss: 1.6036 - val_accuracy: 0.4393 - 4s/epoch - 32ms/step Epoch 24/30 121/121 - 4s - loss: 1.6582 - accuracy: 0.3787 - val_loss: 1.6366 - val_accuracy: 0.4322 - 4s/epoch - 33ms/step Epoch 25/30 121/121 - 4s - loss: 1.6642 - accuracy: 0.3733 - val_loss: 1.6247 - val_accuracy: 0.4369 - 4s/epoch - 34ms/step Epoch 26/30 121/121 - 4s - loss: 1.6388 - accuracy: 0.3832 - val_loss: 1.6590 - val_accuracy: 0.4299 - 4s/epoch - 33ms/step Epoch 27/30 121/121 - 4s - loss: 1.6508 - accuracy: 0.3754 - val_loss: 1.6364 - val_accuracy: 0.4369 - 4s/epoch - 32ms/step Epoch 28/30 121/121 - 4s - loss: 1.6488 - accuracy: 0.3878 - val_loss: 1.6369 - val_accuracy: 0.4299 - 4s/epoch - 33ms/step Epoch 29/30 121/121 - 4s - loss: 1.6425 - accuracy: 0.3881 - val_loss: 1.6199 - val_accuracy: 0.4416 - 4s/epoch - 33ms/step Epoch 30/30 121/121 - 4s - loss: 1.6291 - accuracy: 0.3949 - val_loss: 1.7153 - val_accuracy: 0.4439 - 4s/epoch - 33ms/step
plt.plot(history8.history['accuracy'])
plt.plot(history8.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model8.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 1.8101 - accuracy: 0.4232 - 155ms/epoch - 10ms/step
y_pred=model8.predict(X_test_normalized)
15/15 [==============================] - 0s 9ms/step
# Obtaining the categorical values from y_test_encoded and y_pred
y_pred_arg=np.argmax(y_pred,axis=1)
y_test_arg=np.argmax(y_test_encoded,axis=1)
# Plotting the Confusion Matrix using confusion matrix() function which is also predefined tensorflow module
confusion_matrix = tf.math.confusion_matrix(y_test_arg,y_pred_arg)
f, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(
confusion_matrix,
annot=True,
linewidths=.4,
fmt="d",
square=True,
ax=ax
)
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model9 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model9.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model9.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model9.add(MaxPooling2D((2, 2), padding = 'same'))
model9.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model9.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model9.add(MaxPooling2D((2, 2), padding = 'same'))
model9.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model9.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model9.add(Dense(16, activation='relu'))
model9.add(Dropout(0.3))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model9.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model9.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model9.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 128, 128, 64) 1792
conv2d_1 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d (MaxPooling2 (None, 64, 64, 32) 0
D)
conv2d_2 (Conv2D) (None, 64, 64, 32) 9248
conv2d_3 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_1 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization (Batch (None, 32, 32, 32) 128
Normalization)
flatten (Flatten) (None, 32768) 0
dense (Dense) (None, 16) 524304
dropout (Dropout) (None, 16) 0
dense_1 (Dense) (None, 12) 204
=================================================================
Total params: 563388 (2.15 MB)
Trainable params: 563324 (2.15 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
history9 = model9.fit(
X_train_normalized, y_train_encoded,
epochs=30,
validation_data=(X_val_normalized,y_val_encoded),
batch_size=32,
verbose=2)
Epoch 1/30 121/121 - 7s - loss: 2.2389 - accuracy: 0.2345 - val_loss: 2.4250 - val_accuracy: 0.1379 - 7s/epoch - 56ms/step Epoch 2/30 121/121 - 5s - loss: 1.8594 - accuracy: 0.3540 - val_loss: 2.3103 - val_accuracy: 0.2290 - 5s/epoch - 38ms/step Epoch 3/30 121/121 - 4s - loss: 1.5119 - accuracy: 0.4669 - val_loss: 2.0657 - val_accuracy: 0.2266 - 4s/epoch - 35ms/step Epoch 4/30 121/121 - 5s - loss: 1.3546 - accuracy: 0.5233 - val_loss: 2.7100 - val_accuracy: 0.2897 - 5s/epoch - 39ms/step Epoch 5/30 121/121 - 5s - loss: 1.2641 - accuracy: 0.5402 - val_loss: 2.1385 - val_accuracy: 0.3855 - 5s/epoch - 42ms/step Epoch 6/30 121/121 - 5s - loss: 1.1213 - accuracy: 0.5984 - val_loss: 2.9967 - val_accuracy: 0.2056 - 5s/epoch - 41ms/step Epoch 7/30 121/121 - 4s - loss: 1.0643 - accuracy: 0.6119 - val_loss: 1.0884 - val_accuracy: 0.6379 - 4s/epoch - 35ms/step Epoch 8/30 121/121 - 4s - loss: 0.9526 - accuracy: 0.6532 - val_loss: 2.9286 - val_accuracy: 0.4953 - 4s/epoch - 33ms/step Epoch 9/30 121/121 - 4s - loss: 0.8522 - accuracy: 0.6914 - val_loss: 1.2920 - val_accuracy: 0.5935 - 4s/epoch - 34ms/step Epoch 10/30 121/121 - 4s - loss: 0.8084 - accuracy: 0.6982 - val_loss: 1.1995 - val_accuracy: 0.6215 - 4s/epoch - 34ms/step Epoch 11/30 121/121 - 4s - loss: 0.7568 - accuracy: 0.7193 - val_loss: 1.1355 - val_accuracy: 0.6589 - 4s/epoch - 33ms/step Epoch 12/30 121/121 - 4s - loss: 0.7497 - accuracy: 0.7219 - val_loss: 1.0236 - val_accuracy: 0.6682 - 4s/epoch - 32ms/step Epoch 13/30 121/121 - 4s - loss: 0.6690 - accuracy: 0.7463 - val_loss: 2.0394 - val_accuracy: 0.5397 - 4s/epoch - 34ms/step Epoch 14/30 121/121 - 4s - loss: 0.6275 - accuracy: 0.7609 - val_loss: 1.2165 - val_accuracy: 0.6589 - 4s/epoch - 34ms/step Epoch 15/30 121/121 - 4s - loss: 0.6156 - accuracy: 0.7697 - val_loss: 3.2296 - val_accuracy: 0.3411 - 4s/epoch - 32ms/step Epoch 16/30 121/121 - 4s - loss: 0.6210 - accuracy: 0.7655 - val_loss: 1.0700 - val_accuracy: 0.6869 - 4s/epoch - 32ms/step Epoch 17/30 121/121 - 4s - loss: 0.5703 - accuracy: 0.7767 - val_loss: 1.1539 - val_accuracy: 0.6916 - 4s/epoch - 37ms/step Epoch 18/30 121/121 - 4s - loss: 0.5327 - accuracy: 0.7931 - val_loss: 1.6504 - val_accuracy: 0.6519 - 4s/epoch - 34ms/step Epoch 19/30 121/121 - 4s - loss: 0.5198 - accuracy: 0.8004 - val_loss: 2.2241 - val_accuracy: 0.5888 - 4s/epoch - 34ms/step Epoch 20/30 121/121 - 4s - loss: 0.5267 - accuracy: 0.7970 - val_loss: 1.3839 - val_accuracy: 0.6729 - 4s/epoch - 34ms/step Epoch 21/30 121/121 - 4s - loss: 0.5000 - accuracy: 0.8037 - val_loss: 2.1263 - val_accuracy: 0.5140 - 4s/epoch - 35ms/step Epoch 22/30 121/121 - 4s - loss: 0.5177 - accuracy: 0.8017 - val_loss: 1.7792 - val_accuracy: 0.6285 - 4s/epoch - 34ms/step Epoch 23/30 121/121 - 4s - loss: 0.4911 - accuracy: 0.8079 - val_loss: 3.0073 - val_accuracy: 0.4626 - 4s/epoch - 33ms/step Epoch 24/30 121/121 - 4s - loss: 0.4782 - accuracy: 0.8175 - val_loss: 1.8606 - val_accuracy: 0.6215 - 4s/epoch - 33ms/step Epoch 25/30 121/121 - 4s - loss: 0.4756 - accuracy: 0.8149 - val_loss: 1.6411 - val_accuracy: 0.6542 - 4s/epoch - 36ms/step Epoch 26/30 121/121 - 4s - loss: 0.4248 - accuracy: 0.8303 - val_loss: 1.7797 - val_accuracy: 0.6355 - 4s/epoch - 34ms/step Epoch 27/30 121/121 - 4s - loss: 0.4658 - accuracy: 0.8274 - val_loss: 1.8125 - val_accuracy: 0.6542 - 4s/epoch - 35ms/step Epoch 28/30 121/121 - 4s - loss: 0.4123 - accuracy: 0.8446 - val_loss: 1.7110 - val_accuracy: 0.6355 - 4s/epoch - 35ms/step Epoch 29/30 121/121 - 4s - loss: 0.4133 - accuracy: 0.8425 - val_loss: 2.0162 - val_accuracy: 0.6332 - 4s/epoch - 35ms/step Epoch 30/30 121/121 - 4s - loss: 0.4089 - accuracy: 0.8420 - val_loss: 1.4466 - val_accuracy: 0.6659 - 4s/epoch - 34ms/step
plt.plot(history9.history['accuracy'])
plt.plot(history9.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model9.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 1.5982 - accuracy: 0.6168 - 257ms/epoch - 17ms/step
y_pred_prob = model9.predict(X_test_normalized)
15/15 [==============================] - 0s 14ms/step
# Determine the threshold based on ROC AUC
thresholds = []
for i in range(12): # Change 12 to the actual number of classes
roc_auc = roc_auc_score(y_test_encoded[:, i], y_pred_prob[:, i])
fpr, tpr, threshold = roc_curve(y_test_encoded[:, i], y_pred_prob[:, i])
best_threshold_index = np.argmax(tpr - fpr)
best_threshold = threshold[best_threshold_index]
thresholds.append(best_threshold)
# Use the determined thresholds for prediction
y_pred_binary = np.zeros_like(y_pred_prob)
for i in range(12): # Change 12 to the actual number of classes
y_pred_binary[:, i] = (y_pred_prob[:, i] > thresholds[i]).astype(int)
# Obtain the accuracy using the new predicted binary values
accuracy_roc_auc = accuracy_score(y_test_encoded, y_pred_binary)
print(f"Test Accuracy (ROC AUC Threshold): {accuracy_roc_auc}")
Test Accuracy (ROC AUC Threshold): 0.3136842105263158
# Plotting the Confusion Matrix using confusion matrix() function
conf_matrix = confusion_matrix(
y_test_encoded.argmax(axis=1),
y_pred_binary.argmax(axis=1)
)
# Display the confusion matrix using seaborn
plt.figure(figsize=(10, 8))
sns.heatmap(conf_matrix, annot=True, fmt="d", xticklabels=np.arange(12), yticklabels=np.arange(12))
plt.title('Confusion Matrix')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()
backend.clear_session()
# Fixing the seed for random number generators
import random
np.random.seed(42)
random.seed(42)
tf.random.set_seed(42)
# Intializing a sequential model
model10 = Sequential()
# Adding first conv layer with 64 filters and kernel size 3x3 , padding 'same' provides the output size same as the input size
# Input_shape denotes input image dimension of images
model10.add(Conv2D(64, (3, 3), activation='relu', padding="same", input_shape=(128, 128, 3)))
model10.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
# Adding max pooling to reduce the size of output of first conv layer
model10.add(MaxPooling2D((2, 2), padding = 'same'))
model10.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model10.add(Conv2D(32, (3, 3), activation='relu', padding="same"))
model10.add(MaxPooling2D((2, 2), padding = 'same'))
model10.add(BatchNormalization())
# flattening the output of the conv layer after max pooling to make it ready for creating dense connections
model10.add(Flatten())
# Adding a fully connected dense layer with 100 neurons
model10.add(Dense(32, activation='relu'))
model10.add(Dropout(0.5))
# Adding the output layer with 10 neurons and activation functions as softmax since this is a multi-class classification problem
model10.add(Dense(12, activation='softmax'))
#using Adam optimizer
opt=Adam()
# Compile model
model10.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# Generating the summary of the model
model10.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_4 (Conv2D) (None, 128, 128, 64) 1792
conv2d_5 (Conv2D) (None, 128, 128, 32) 18464
max_pooling2d_2 (MaxPoolin (None, 64, 64, 32) 0
g2D)
conv2d_6 (Conv2D) (None, 64, 64, 32) 9248
conv2d_7 (Conv2D) (None, 64, 64, 32) 9248
max_pooling2d_3 (MaxPoolin (None, 32, 32, 32) 0
g2D)
batch_normalization_1 (Bat (None, 32, 32, 32) 128
chNormalization)
flatten_1 (Flatten) (None, 32768) 0
dense_2 (Dense) (None, 32) 1048608
dropout_1 (Dropout) (None, 32) 0
dense_3 (Dense) (None, 12) 396
=================================================================
Total params: 1087884 (4.15 MB)
Trainable params: 1087820 (4.15 MB)
Non-trainable params: 64 (256.00 Byte)
_________________________________________________________________
# Code to monitor val_accuracy
reduce_lr = ReduceLROnPlateau(monitor='val_accuracy',
patience=3,
verbose=1,
factor=0.5,
min_lr=0.00001)
history10 = model10.fit(
X_train_normalized, y_train_encoded,
epochs=30,
validation_data=(X_val_normalized, y_val_encoded),
batch_size=32,
verbose=2,
callbacks=[reduce_lr]
)
Epoch 1/30 121/121 - 7s - loss: 2.1359 - accuracy: 0.2789 - val_loss: 2.4063 - val_accuracy: 0.1822 - lr: 0.0010 - 7s/epoch - 54ms/step Epoch 2/30 121/121 - 4s - loss: 1.7066 - accuracy: 0.4167 - val_loss: 2.3228 - val_accuracy: 0.1729 - lr: 0.0010 - 4s/epoch - 35ms/step Epoch 3/30 121/121 - 4s - loss: 1.4879 - accuracy: 0.4926 - val_loss: 1.9852 - val_accuracy: 0.4252 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 4/30 121/121 - 4s - loss: 1.4029 - accuracy: 0.5147 - val_loss: 1.6024 - val_accuracy: 0.5117 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 5/30 121/121 - 4s - loss: 1.2788 - accuracy: 0.5656 - val_loss: 1.5945 - val_accuracy: 0.4907 - lr: 0.0010 - 4s/epoch - 35ms/step Epoch 6/30 121/121 - 4s - loss: 1.1737 - accuracy: 0.5841 - val_loss: 1.1178 - val_accuracy: 0.6402 - lr: 0.0010 - 4s/epoch - 35ms/step Epoch 7/30 121/121 - 4s - loss: 1.0978 - accuracy: 0.6059 - val_loss: 1.2592 - val_accuracy: 0.6051 - lr: 0.0010 - 4s/epoch - 33ms/step Epoch 8/30 121/121 - 4s - loss: 1.0196 - accuracy: 0.6278 - val_loss: 1.2044 - val_accuracy: 0.6425 - lr: 0.0010 - 4s/epoch - 33ms/step Epoch 9/30 121/121 - 4s - loss: 0.9620 - accuracy: 0.6592 - val_loss: 2.2442 - val_accuracy: 0.4556 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 10/30 121/121 - 4s - loss: 0.9417 - accuracy: 0.6543 - val_loss: 1.0824 - val_accuracy: 0.6659 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 11/30 121/121 - 4s - loss: 0.8503 - accuracy: 0.6985 - val_loss: 1.2646 - val_accuracy: 0.6145 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 12/30 121/121 - 4s - loss: 0.8313 - accuracy: 0.6959 - val_loss: 1.9335 - val_accuracy: 0.5654 - lr: 0.0010 - 4s/epoch - 33ms/step Epoch 13/30 Epoch 13: ReduceLROnPlateau reducing learning rate to 0.0005000000237487257. 121/121 - 4s - loss: 0.7808 - accuracy: 0.7073 - val_loss: 1.6651 - val_accuracy: 0.6285 - lr: 0.0010 - 4s/epoch - 34ms/step Epoch 14/30 121/121 - 4s - loss: 0.6590 - accuracy: 0.7528 - val_loss: 1.2815 - val_accuracy: 0.6121 - lr: 5.0000e-04 - 4s/epoch - 33ms/step Epoch 15/30 121/121 - 4s - loss: 0.6320 - accuracy: 0.7624 - val_loss: 1.0678 - val_accuracy: 0.6986 - lr: 5.0000e-04 - 4s/epoch - 33ms/step Epoch 16/30 121/121 - 4s - loss: 0.6340 - accuracy: 0.7598 - val_loss: 0.9532 - val_accuracy: 0.7173 - lr: 5.0000e-04 - 4s/epoch - 34ms/step Epoch 17/30 121/121 - 4s - loss: 0.5672 - accuracy: 0.7837 - val_loss: 1.1993 - val_accuracy: 0.6449 - lr: 5.0000e-04 - 4s/epoch - 34ms/step Epoch 18/30 121/121 - 4s - loss: 0.5615 - accuracy: 0.7814 - val_loss: 1.2611 - val_accuracy: 0.7150 - lr: 5.0000e-04 - 4s/epoch - 33ms/step Epoch 19/30 Epoch 19: ReduceLROnPlateau reducing learning rate to 0.0002500000118743628. 121/121 - 4s - loss: 0.5449 - accuracy: 0.7892 - val_loss: 1.2877 - val_accuracy: 0.6916 - lr: 5.0000e-04 - 4s/epoch - 36ms/step Epoch 20/30 121/121 - 4s - loss: 0.5133 - accuracy: 0.7985 - val_loss: 1.0661 - val_accuracy: 0.7336 - lr: 2.5000e-04 - 4s/epoch - 35ms/step Epoch 21/30 121/121 - 4s - loss: 0.4752 - accuracy: 0.8131 - val_loss: 1.1244 - val_accuracy: 0.7453 - lr: 2.5000e-04 - 4s/epoch - 34ms/step Epoch 22/30 121/121 - 4s - loss: 0.4642 - accuracy: 0.8123 - val_loss: 1.1950 - val_accuracy: 0.7150 - lr: 2.5000e-04 - 4s/epoch - 34ms/step Epoch 23/30 121/121 - 4s - loss: 0.4611 - accuracy: 0.8225 - val_loss: 1.1081 - val_accuracy: 0.7336 - lr: 2.5000e-04 - 4s/epoch - 33ms/step Epoch 24/30 Epoch 24: ReduceLROnPlateau reducing learning rate to 0.0001250000059371814. 121/121 - 4s - loss: 0.4535 - accuracy: 0.8245 - val_loss: 1.0947 - val_accuracy: 0.7196 - lr: 2.5000e-04 - 4s/epoch - 34ms/step Epoch 25/30 121/121 - 4s - loss: 0.4328 - accuracy: 0.8349 - val_loss: 1.1109 - val_accuracy: 0.7407 - lr: 1.2500e-04 - 4s/epoch - 35ms/step Epoch 26/30 121/121 - 4s - loss: 0.3989 - accuracy: 0.8409 - val_loss: 1.1505 - val_accuracy: 0.7523 - lr: 1.2500e-04 - 4s/epoch - 36ms/step Epoch 27/30 121/121 - 4s - loss: 0.4314 - accuracy: 0.8264 - val_loss: 1.1505 - val_accuracy: 0.7407 - lr: 1.2500e-04 - 4s/epoch - 33ms/step Epoch 28/30 121/121 - 4s - loss: 0.4218 - accuracy: 0.8316 - val_loss: 1.1382 - val_accuracy: 0.7430 - lr: 1.2500e-04 - 4s/epoch - 34ms/step Epoch 29/30 Epoch 29: ReduceLROnPlateau reducing learning rate to 6.25000029685907e-05. 121/121 - 4s - loss: 0.3971 - accuracy: 0.8438 - val_loss: 1.1789 - val_accuracy: 0.7290 - lr: 1.2500e-04 - 4s/epoch - 35ms/step Epoch 30/30 121/121 - 4s - loss: 0.3798 - accuracy: 0.8440 - val_loss: 1.1647 - val_accuracy: 0.7430 - lr: 6.2500e-05 - 4s/epoch - 33ms/step
plt.plot(history10.history['accuracy'])
plt.plot(history10.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()
accuracy = model10.evaluate(X_test_normalized, y_test_encoded, verbose=2)
15/15 - 0s - loss: 1.1067 - accuracy: 0.7179 - 256ms/epoch - 17ms/step
y_pred_prob=model10.predict(X_test_normalized)
15/15 [==============================] - 0s 13ms/step
# Determine the threshold based on ROC AUC
thresholds = []
for i in range(12): # Change 12 to the actual number of classes
roc_auc = roc_auc_score(y_test_encoded[:, i], y_pred_prob[:, i])
fpr, tpr, threshold = roc_curve(y_test_encoded[:, i], y_pred_prob[:, i])
best_threshold_index = np.argmax(tpr - fpr)
best_threshold = threshold[best_threshold_index]
thresholds.append(best_threshold)
# Use the determined thresholds for prediction
y_pred_binary = np.zeros_like(y_pred_prob)
for i in range(12): # Change 12 to the actual number of classes
y_pred_binary[:, i] = (y_pred_prob[:, i] > thresholds[i]).astype(int)
# Obtain the accuracy using the new predicted binary values
accuracy_roc_auc = accuracy_score(y_test_encoded, y_pred_binary)
print(f"Test Accuracy (ROC AUC Threshold): {accuracy_roc_auc}")
Test Accuracy (ROC AUC Threshold): 0.45473684210526316
# Plotting the Confusion Matrix using confusion matrix() function
conf_matrix = confusion_matrix(
y_test_encoded.argmax(axis=1),
y_pred_binary.argmax(axis=1)
)
# Display the confusion matrix using seaborn
plt.figure(figsize=(10, 8))
sns.heatmap(conf_matrix, annot=True, fmt="d", xticklabels=np.arange(12), yticklabels=np.arange(12))
plt.title('Confusion Matrix')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.show()
pd.DataFrame({'Models':['Model1', 'Model2', 'Model3', 'Model4', 'Model5', 'Model6', 'Model7', 'Model8', 'Model9','Model10'],
'Train Accuracy':['43%','87%','59%','42%','51%','29%','14%','39%','84%','84%'],
'Validation Accuracy':['49%','60%','45%','23%','47%','43%','14%','44%','67%','74%'],
'Test Accuracy':['50%','61%','46%','24%','44%','44%','14%','42%','62%','72%']
})
| Models | Train Accuracy | Validation Accuracy | Test Accuracy | |
|---|---|---|---|---|
| 0 | Model1 | 43% | 49% | 50% |
| 1 | Model2 | 87% | 60% | 61% |
| 2 | Model3 | 59% | 45% | 46% |
| 3 | Model4 | 42% | 23% | 24% |
| 4 | Model5 | 51% | 47% | 44% |
| 5 | Model6 | 29% | 43% | 44% |
| 6 | Model7 | 14% | 14% | 14% |
| 7 | Model8 | 39% | 44% | 42% |
| 8 | Model9 | 84% | 67% | 62% |
| 9 | Model10 | 84% | 74% | 72% |
rows = 2
cols = 5
# Create a figure with the specified number of rows and columns
fig, axes = plt.subplots(rows, cols, figsize=(15, 10))
# Flatten the axes for easy iteration
axes = axes.flatten()
for i in range(rows * cols):
random_index = np.random.randint(0, len(X_test_normalized))
ax = axes[i]
# Display the image
ax.imshow(X_test[random_index])
# Get the predicted label, probability, and true label
pred_label = np.argmax(y_pred_binary[random_index])
pred_proba = y_pred_prob[random_index, pred_label]
true_label = np.argmax(y_test_encoded[random_index])
# Set the title with actual, predicted, and probability information
ax.set_title(f"Actual: {true_label}\nPredicted: {pred_label}\nProbability: {pred_proba:.3f}", fontsize=10)
# Remove the x and y ticks for better visualization
ax.set_xticks([])
ax.set_yticks([])
# Adjust layout for better spacing
plt.tight_layout()
plt.show()
